ABSTRACT
Heretofore, we focused our attention primarily on models for data in which the outcome variables are continuous. Indeed, we have been even more specific and dealt almost exclusively with models resting on the assumption that errors are normally distributed. However, in many applications, the outcome variable of interest is categorical rather than continuous. For example, a researcher may be interested in predicting whether an incoming freshman is likely to graduate from college in 4 years, using high school grade point average and admissions test scores as the independent variables. Here, the outcome is a dichotomous variable: graduation in 4 years (yes or no). Likewise, consider research conducted by a linguist who interviewed terminally ill patients and wants to compare the number of times those patients use the words death and dying during the interviews. The number of times that each word appears, when compared to the many thousands of words contained in the interviews, is likely to be very small, if not zero for some people.
